Agents and workflows

Use this page when the main problem is routing, controller loops, tool execution, memory boundaries, or orchestration.

Manage LLM memory boundaries

Use this when memory or persistent context can affect workflow behavior.

Use with:
Open guide →
Run the architecture boundary review

Use this when boundaries, ownership, dependencies, or architecture direction need review.

Use with:
Open guide →
Run the standards-backed implementation review

Use this when the workflow must be checked against official documentation.

Use with:
Open guide →
Request Assembly Threat Model

Use this when request construction, source selection, or context assembly is the main risk.

Use with:
Read article →

When to use agents and workflows

Use this page when the work spans more than one prompt or one assistant reply.
Use it when the system includes an orchestrator, controller loop, retrieval step, memory layer, external tools, or multi-step review path.
Use it when you need to review trust boundaries, authority transfer, policy enforcement, or workflow routing.

Before you start

Know which component owns routing, tool calls, memory, retrieval, and final output.
Know which data sources and tools the workflow can access.
Know whether the issue is architecture, policy enforcement, request assembly, or runtime behavior.

Prompt engineering
The main problem is the prompt text or reusable instruction pattern.
AI assistants
The main problem is assistant-product behavior inside ChatGPT, Claude, or Gemini.
API integrations and tools
The main problem is SDK/API wiring, tool calls, or external-system integration.